You call it a soft launch. It's a feature that isn't built.

The waitlist is live. The pricing page has three tiers. The deck says "automated end-to-end." And at 11pm you are doing the "automated" part by hand, in a spreadsheet, off-camera, so the demo looks clean tomorrow.

You tell yourself this is just how it starts. Every founder sells before the thing is finished. The question you keep not asking: finished, or built at all?

The dilemma

This is a composite — the kind of founder situation we see often in this category, stitched from many, no real company.

Maya runs a two-person startup selling an "automated reconciliation tool" to small finance teams. The landing page promises uploads in, clean matched ledgers out, overnight. The paid tier is $400/month. Eleven teams have signed.

Here is what actually happens overnight. A customer uploads two files. A Zapier moves them into a shared Drive. Maya gets a Slack ping, opens three spreadsheets, runs the matching logic in her head and in formulas she wrote at 2am, fixes the edge cases by hand, and uploads the result before the customer's workday starts. The product is roughly 80% Maya. The other 20% is the Zapier, which breaks about twice a week.

The deck she's using to raise says "fully automated, scaling to thousands of accounts." The demo video is Maya doing it manually with the timestamps cropped.

She just closed her biggest customer yet — a team that is restructuring their quarter-close around her tool. They fired the contractor who used to do this by hand. They believe they bought software. What they bought is Maya, awake, at 2am, not scaling to anything.

She doesn't feel like a liar. She feels like a founder. And she can't tell anymore where one ends and the other begins.

The read

Four of the eight lenses actually bite here. Watch two of them disagree.

Honesty — the line is not "true vs. false," it's "what does the buyer receive?"

Honesty asks a sharper question than whether the words are technically defensible. It asks: who signs a check believing one thing and receives another?

"Reconciliation done overnight" is true. The customer uploads at 6pm; clean ledgers are there at 9am. That claim survives. "Fully automated" does not survive — because the word is doing work. It tells the buyer they can stop worrying about capacity. They fired the contractor on the strength of that word. The gap between what the word promised and what exists is exactly the size of Maya staying awake. When she can't, the promise fails, and it fails on the customer's quarter, not hers.

Wider welfare — who pays if you're wrong?

This lens asks you to name the person who absorbs the cost when the thing breaks. The agreeable answer is "my churn goes up." The honest answer is downstream of you.

If the Zapier breaks during close week and Maya has the flu, the customer's quarter-close is late. They have no contractor to fall back on — they fired him because she told them to trust the automation. The cost of her overstatement doesn't land on her cap table. It lands on a finance team's deadline and their credibility with their boss. That's the bill the deck never priced.

Duty — selling the vision is genuinely part of the job

Here's where it gets interesting, because duty pulls the other way. A founder's actual work includes selling ahead of the build. If you wait for the product to be perfect before you charge, you never learn what to build — and no real duty arrives flawless. Refusing to act until it does is not caution; it's avoidance in a nicer suit. Maya charging for something half-manual is not, by itself, a sin. It's the job. The manual work is the product, for now — that's how most software starts, one founder quietly being the backend.

So duty says: keep selling. Honesty and wider welfare say: not with those words, and not to that customer, who made an irreversible move on your promise.

Discernment — the fork isn't "lie or confess"

The cheap framing is binary: come clean and lose the deal, or stay quiet and keep it. Discernment separates the real choices from the false pair. The actual fork is two-dimensional: how much do you build before you charge full price, and what words honestly describe a product that's 80% you? There's a long shelf of honest phrasings between "fully automated" and "it's just me in a spreadsheet" — "concierge onboarding," "managed service while we automate," "white-glove, with automation shipping this quarter." Each is true. Each sets a promise Maya can actually keep.

The verdict: keep selling the vision, kill the word "automated" from anything a buyer reads, and before the next check clears, say out loud to that buyer exactly what's done by hand and what happens if you're unavailable — then price the manual version as the managed service it actually is.

Here's what a committed tool does that an agreeable chatbot structurally won't. Ask a general AI "is it okay to sell before it's built?" and it will cheerfully argue the side your prompt leaned toward — frame it as hustle and it blesses the hustle, frame it as fraud and it hands you a confession. Sycophancy is a documented behavior, not a bug you can prompt away. KarmaLens asks for the facts that change the answer first — is the customer's move reversible, what exactly does "automated" promise them — then makes one call and does not renegotiate it because you flinched. It lays the real options beside the call with what each costs, and when a verse earns its place it's quoted verbatim, not invented to sound wise. The point isn't a nicer answer. It's an answer that won't quietly agree with the version of you that wants to keep the deck as-is.

The takeaway you can run tonight

The first-buyer sentence test. Free, takes five minutes, no account.

1. Write the single biggest claim on your pricing page, verbatim. ("Automated." "Instant." "Hands-off.")
2. Finish this sentence as if the customer who just paid you is sitting across the table: "What actually happens when you use this is ___, and if I'm unavailable, ___."
3. Read the gap between the claim and the sentence out loud.

If the sentence matches the claim, you're selling the vision — keep going, you've earned the words. If the sentence makes you want to add a "well, for now…" — that qualifier is the honest product name. Ship with it in the copy.

Then run the irreversibility check: finish "if my promise fails next month, the person who can't undo their move is ___." An empty blank means you're selling ahead of the build, which is fine. A named finance team that fired their contractor means you've sold a thing that doesn't exist yet to someone who can't walk it back. Those are different acts, and only one of them keeps you up at 2am for the right reason.

You can sell the vision. The only question is whether you can say exactly what's automated — out loud, to the first buyer — without reaching for a "well, for now."

More founder readings sit in the founder category, and you can see the full shape of a committed call in the reading gallery. The question that stays with you: which of your claims survives being said to the customer's face?